Transcribe a piano recording to sheet music

Updated July 2026

A piano hands a transcriber the worst version of every problem at once: ten notes sounding together, a pedal that hides where each one ends, and a performance that is deliberately not on a click. The work is not finding pitches — it is deciding which of the overlapping sounds were separate notes.

Polyphony is the whole problem

A melodic line is one pitch at a time. A piano is a stack: a left-hand tenth under a four-note voicing under a doubled melody. Those notes share harmonics. The overtone series of a low C already contains a C an octave up, a G above that, another C above that — and every one of those partials looks like a note somebody might have played.

That is why piano is routed to a dedicated polyphonic model rather than the general one. It is also why octave errors are the first thing to check in a piano transcription: a doubled octave collapsed into one note, or a phantom note an octave above a loud bass, are the characteristic mistakes.

The pedal is why note lengths are the weak spot

Onsets survive the sustain pedal. Releases do not. With the pedal down a chord rings straight through the next one, and the boundary between where a note stopped being played and where it stopped being audible is genuinely ambiguous in the signal — not just hard to measure.

The practical consequence: read the note lengths before you read the pitches. Values held too long across a pedalled passage are the expected failure, and they take seconds to trim once you know to look.

Rubato is the reason you recorded it

Piano is where a fixed grid does the most damage. Assume a constant tempo and a phrase that leans into a cadence comes back as wrong rhythms — quintuplets and dotted figures nobody played, printed with total confidence.

Tempo, downbeat and meter are detected before the notes are fitted, and the tempo map is allowed to move. The performance is written against the tempo it was actually played at, which is the only way the note values stay recognisable.

Clefs, key signature, spelling

Each part is engraved on the clef that suits its register, so a bass-register line lands on bass clef instead of trailing ledger lines. The key is detected and printed as a key signature, and accidentals appear only where a note departs from it — not a sharp stamped on every F because the tool only knows MIDI note numbers.

Spelling follows the key as well. In A♭ major the black key between C and D is D♭; in D major the same key is C♯. That one choice is most of what makes a transcription look like music instead of like data.

If the piano is inside a mix, separate first

To transcribe a piano recording to sheet music when the piano is buried in a finished track, do not point a transcriber at the mixdown and hope. Split the song into stems — vocals, drums, bass, guitar, keys — and transcribe the keys stem on its own. That ordering is the single biggest accuracy win available, and it matters more for piano than for almost anything, because a piano covers the entire frequency range and everything else in the mix overlaps it somewhere.

Edit it, then export it

No transcription of a real recording is perfect, which is exactly why the result is editable. It opens as a multitrack MIDI arrangement, one lane per part, where you fix a wrong note or trim a held value; the score re-engraves from that same edit and the exported MIDI carries it.

Exports are MIDI per part, or a Logic or Ableton session with stems on named tracks, per-part MIDI, the tempo map and a chord guide. Final editorial engraving still belongs in a notation editor — this gets you a correct, readable score to take there.

Related

• AI music transcription
• Turn a recording into sheet music
• Turn a recording into MIDI
• Split a song into stems
• Turn a mix into a session
• All guides